Compute Bidirectional Correlation

Computes the mean of two correlations: cor(t(A_w1) %% K, B_w2) and cor(A_w1, K %% B_w2), where A_w1 and B_w2 are cell scores derived from PCA matrices multiplied by skrCCA weights for the corresponding cell types, and K is the kernel matrix between the two cell types.

Usage

computeBidirCorrelation(
  object,
  calculationMode = "perSlide",
  normalize_K = c("row_or_col", "sinkhorn_knopp", "none"),
  filter_kernel = TRUE,
  K_row_sum_cutoff = 0.05,
  K_col_sum_cutoff = 0.05
)

# S4 method for class 'CoPro'
computeBidirCorrelation(
  object,
  calculationMode = "perSlide",
  normalize_K = c("row_or_col", "sinkhorn_knopp", "none"),
  filter_kernel = TRUE,
  K_row_sum_cutoff = 0.05,
  K_col_sum_cutoff = 0.05
)

# S4 method for class 'CoProMulti'
computeBidirCorrelation(
  object,
  calculationMode = "perSlide",
  normalize_K = c("row_or_col", "sinkhorn_knopp", "none"),
  filter_kernel = TRUE,
  K_row_sum_cutoff = 0.05,
  K_col_sum_cutoff = 0.05
)

Arguments

object

A CoPro or CoProMulti object containing skrCCA results, PCA results, and kernel matrices.

calculationMode

(CoProMulti only) "perSlide" or "aggregate". Ignored for single-slide objects.

normalize_K

whether to normalize the kernel matrix, one of "none", "sinkhorn_knopp", "row_or_col"

filter_kernel

whether to filter the kernel matrix, default is TRUE

K_row_sum_cutoff

Numeric; cutoff for row sums when normalizing kernel matrix. Default 5e-3.

K_col_sum_cutoff

Numeric; cutoff for column sums when normalizing kernel matrix. Default 5e-3.

Value

The input object with @bidirCorrelation populated (a list keyed by sigma names). For single-slide objects, each entry is a data frame with columns sigmaValues, cellType1, cellType2, CC_index, bidirCorrelation. For multi-slide objects, the columns depend on calculationMode.

Details

The results are stored in the slot bidirCorrelation.

For multi-slide (CoProMulti) objects, results can be computed per slide (default, returns a data frame with a slideID column) or aggregated across slides (returns a data frame with an aggregateCorrelation column).

Examples

# Assuming `obj` is a prepared CoProSingle with PCA, kernels, and skrCCA:
# obj <- computeBidirCorrelation(obj)

# For CoProMulti per-slide results:
# objm <- computeBidirCorrelation(objm, calculationMode = "perSlide")

# For CoProMulti aggregate results:
# objm <- computeBidirCorrelation(objm, calculationMode = "aggregate")